Compute precision for output scalar

python/cudf/cudf/_lib/reduce.pyx

@@ -1,7 +1,8 @@
-# Copyright (c) 2020, NVIDIA CORPORATION.
+# Copyright (c) 2020-2021, NVIDIA CORPORATION.
 
 import cudf
 from cudf.utils.dtypes import is_decimal_dtype
+from cudf.core.dtypes import Decimal64Dtype
 from cudf._lib.cpp.reduce cimport cpp_reduce, cpp_scan, scan_type, cpp_minmax
 from cudf._lib.cpp.scalar.scalar cimport scalar
 from cudf._lib.cpp.types cimport data_type, type_id
@@ -68,7 +69,14 @@ def reduce(reduction_op, Column incol, dtype=None, **kwargs):
             c_out_dtype
         ))
 
-    py_result = DeviceScalar.from_unique_ptr(move(c_result), dtype=col_dtype)
+    if c_result.get()[0].type().id() == libcudf_types.type_id.DECIMAL64:
+        scale = -c_result.get()[0].type().scale()
+        precision = _reduce_precision(col_dtype, reduction_op, len(incol))
+        py_result = DeviceScalar.from_unique_ptr(
+            move(c_result), dtype=Decimal64Dtype(precision, scale)
+        )
+    else:
+        py_result = DeviceScalar.from_unique_ptr(move(c_result))
     return py_result.value
 
 
@@ -131,3 +139,24 @@ def minmax(Column incol):
     py_result_max = DeviceScalar.from_unique_ptr(move(c_result.second))
 
     return cudf.Scalar(py_result_min), cudf.Scalar(py_result_max)
+
+
+def _reduce_precision(dtype, op, nrows):
+    """
+    Returns the result precision when performing the reduce
+    operation `op` for the given dtype and column size.
+
+    See: https://docs.microsoft.com/en-us/sql/t-sql/data-types/precision-scale-and-length-transact-sql
+    """  # noqa: E501
+    p = dtype.precision
+    if op in ("min", "max"):
+        new_p = p
+    elif op == "sum":
+        new_p = p + nrows - 1
+    elif op == "product":
+        new_p = p * nrows + nrows - 1
+    elif op == "sum_of_squares":
+        new_p = 2 * p + nrows
+    else:
+        raise NotImplementedError()
+    return max(min(new_p, Decimal64Dtype.MAX_PRECISION), 0)